On the sensitivity of Cronbach's alpha to fake data